Sclera recognition, a rapidly growing area in biometrics, offers significant potential for secure identity verification. This study examines the effectiveness of Convolutional Neural Networks, CNN,in terms of feature extraction in sclera recognition systems. Neural Networks have exhibits exceptional performance in various image processing applications, predominantly because of its capability to automatically learn discriminative features from any given raw data. We evaluate the suitability of CNN architectures, including DenseNet121, InceptionV3, VGG16, and ResNet for extracting meaningful features from sclera images. By using transfer learning and refinement techniques, we focused to utilize the power of pre-trained CNN models on extensive datasets to increase the performance of sclera recognition methods. Our research advances the state-of-the-art in sclera biometrics by providing insights into the effectiveness of CNN-based feature extraction methods, thereby paving the way for more accurate and reliable identification systems in diverse real-world scenarios.